UNIVERSITY OF CALIFORNIA College of Engineering Department of Electrical Engineering and Computer Sciences

Transcription

1 UNIVERSITY OF CALIFORNIA College of Engineering Department of Electrical Engineering and Computer Sciences E. Alon Final EECS 240 Monday, May 19, 2008 SPRING 2008 You should write your results on the exam sheets only. Partial credit will be given only if you show your work and reasoning clearly. Throughout the exam, you can ignore the r o of any transistors and all capacitors except those explicitly drawn in the diagrams unless the problem states otherwise. Name: SID: Problem 1 / 16 Problem 2 / 16 Problem 3 / 20 Problem 4 / 20 Total / 72 1

2 Problem 1 (16 points) Active Inductors a) (4 pts) As we will explore further in this problem, many of the equalizers discussed in class make use of so-called active inductors. For what values of R eq and L eq will the Z in s of the two circuits shown below exactly equal each other? You should provide your answer in terms of the g m of the transistor and ω z = (R z C z ) -1. You should assume that the unity-gain buffer in the active inductor is ideal (i.e., has infinite input impedance and zero output impedance). b) (2 pts) Given your answer to part a), if g m = 500µS, what value of L eq is required for ω z to equal 2π 200MHz? 2

3 c) (4 pts) For the equalizing amplifier shown below, what is the variance of the noise at V out due only to the thermal noise of R L? (i.e., you can ignore the noise from the input transistor) You should provide your answer in terms of kt, C L, L eq, and R L. 3

4 d) (6 pts) What is the noise variance at V out due only to the thermal noise of the load for the active inductor equalizer shown below? You can ignore the noise from R z, but note that there is no longer a buffer isolating R z and C z from the output (i.e., you should include the loading due to R z and C z ). You should provide your answer in terms of kt, γ, ω z, C L, and C z. 4

5 Problem 2 (16 points) Equalizer Design This problem will examine the link shown below, where the channel behaves as a singlepole low-pass filter, and the receiver uses the R1/R2/C1 network to perform equalization. You can assume that the amplifier in the receiver has zero output impedance s ωchan a) (4 pts) Assuming R1 is fixed, write the two equations you would use to set R2 and C1 in order to achieve an equalized 3dB bandwidth of ω bit_rate over a channel with a pole at ω chan. Given these relationships, what is the DC gain of the equalizer? (Hint: what are the pole and zero locations of the equalizer?) 5

6 b) (4 pts) Given your answer to a), if we d like the overall gain (i.e., V out /V rx ) of the equalizer to be equal to A eq over a 3dB bandwidth of ω bit_rate, what is the total gain-bandwidth required of the amplifier? You should provide your answer in terms of A eq, ω chan, and ω bit_rate. 6

7 c) (4 pts) Now let s assume that due to a design error, both the zero and the DC gain of the equalizer are 3 times lower than their expected values. Ignoring the finite bandwidth of the amplifier, sketch the new equalized response in the frequency domain (i.e., V out (jω)/v tx (jω) ). Be sure to label the magnitudes and frequencies of all relevant break-points. d) (4 pts) For the situation in c), how much smaller is the worst-case eye opening in comparison to the case where R1 takes on the correct value? (Hint: Sketch the step response of V out /V tx ) 7

8 Problem 3 (20 points) CML Latch In this problem we will be looking at the CML latch shown below. M1, M2, M3, and M4 all have a channel length of 0.2µm and the same width. The widths of these devices are chosen to achieve a fixed V* of 200mV. M5 and M6 are operated as switches that you can assume are ideal. Unless otherwise noted, you should use the following design and technology parameters: R L = 2kΩ I b = 250µA For V* = 200mV, I DS /W = 25µA/µm. V DD = 1.8V A Vth = 8mV µm A β = 4% µm a) (3 pts) What is the gain from V i to V o while M5 is on and M6 is off? 8

9 b) (4 pts) What is the standard deviation (σ) of the input-referred offset of the latch due only to M1 and M2? Don t forget to include the impact of β mismatch. c) (6 pts) Given your answers to parts a) and b), what is the total σ of the inputreferred offset of the latch? (Note that you can assume that the resistors are perfectly matched, but don t forget to include the impact of M3 and M4.) 9

10 d) (4 pts) Assuming your answer to part c) was σ offset =10mV, how much power would the latch need to dissipate in order to reduce the σ offset to 2mV? As you resize the latch to reduce the offset, you should maintain the same V*=200mV for all of the devices, as well as the same bandwidth and regeneration time constant (remember that the gate capacitors of M3 and M4 will load the output). 10

11 e) (3 pts) Now returning to the original latch design with I b = 250µA and still assuming that the answer to part c) was σ offset =10mV, let s consider the power required to cancel the offset of the latch using the programmable current trim circuit shown below. What is the maximum value of I off+ or I off- required to cancel 3σ (i.e., 30mV) of input-referred offset? 11

12 Problem 4 (20 points) Miscellaneous a) (6 pts) What is the small-signal loop gain T(s) of the common-mode feedback loop shown below? You can assume that the OTA has a finite transconductance of G m, but otherwise is ideal (i.e., it has infinite input and output impedance). 12

13 b) (6 pts) In order to save power by turning on their biasing circuitry for only 10ns every 100us, one of your colleagues suggests augmenting their current mirrors with a sampling switch as shown below. What will likely be the largest source of error in the output current caused by this approach? What design steps could you take to mitigate this error? (Two correctly explained suggestions will receive full credit on this problem.) 13

14 c) (8 pts) Shown below is a two-stage Miller compensated amplifier tied in unitygain feedback along with its bias generation circuit. How can you modify the bias circuit to ensure that the closed-loop bandwidth of the amplifier remains constant across process, voltage, and temperature variations? (Hint: What reference source might be available that is measured in the same units as bandwidth?) To receive full credit on this problem, you should describe the relationships that set the closed-loop bandwidth of your modified circuit. 14

15 15